On 19 May 2015, at 23:09, meekerdb wrote:
On 5/19/2015 11:47 AM, Terren Suydam wrote:
While I applaud IIT because it seems to be the first theory of
consciousness that takes information architecture seriously (and
thus situating theoretical considerations in a holistic rather than
reductionist context) and to make predictions based on that, I
agree with Aaronson's criticisms of it - namely, that IIT predicts
that certain classes of computational systems that we intuitively
would fail to see as conscious get measures of consciousness
potentially higher than for human brains.
One key feature of consciousness as we know it is ongoing
subjective experience. So a question I keep coming back to in my
own thinking is, what kind of information architecture lends itself
to a flow of data, such that if we assume that "consciousness is
how data feels as it's processed", we might imagine it could
correspond to ongoing subjective experience? It seems to me that
such an architecture would have, at a bare minimum, its current
state recursively fed back into itself to be processed in the next
iteration. This happens in a trivial way in any processor chip (or
lookup table AI for that matter). As such, there may be a very
trivial sort of consciousness associated with a processor
or lookup table, but this does not get us anywhere near
understanding the richness of human consciousness.
I think you need to consider what would be the benefit of this
recursion. How could it be naturally selected. Jeff Hawkins idea
is that the brain continually tries to anticipate, at the perceptual
level and even in lower layers of the cerebral cortex.
I think that is the case, and it is the case for the []p & <>t, which
is a form of bet/anticipation. That idea is the base of Helmholtz
theory of perception, and many experience in psychology confirms this
idea.
Then signals that don't match the prediction get broadcast more
widely at the next higher level where they may have been anticipated
by other neurons. At the highest level (he says there are six in
the cortext as I recall) signals spread to language and visual
modules and one "becomes aware of them" or "they spring to mind".
This would have the advantage of directing computational resources
to that which is novel, while leaving familiar things to learned
responses. To this I would add that the novel/conscious experience
is given some value, e.g. emotional weight, which makes it more or
less strongly remembered. And of course it isn't remembered like
recording; it's synopsized in terms of it's connection to other
remembered events. This memory is needed for learning from experience.
OK.
Of course the loop Terren alluded too is built in in the [], and the
usual self-reference brought bu the use of the second recursion
theorem, or the Dx = xx trick. Such loop are the technical base of all
the modal logics of self-reference. The second recursion theorem is
hidden in the proof of Solovay's theorem.
Bruno
Brent
An architecture that supports that richness - the subjective
experience, IOW, of an embodied sensing agent - would involve that
recursion but at a holistic level. The entire system, potentially,
including the system's informational representations of sensory
data (whatever form that took) would be involved in that feedback
loop. So the phi of IIT has a role here, as the processor/lookup
table architecture has a low phi.
What is missing from phi is a measure of recursion - how the
modules of a system feedback in such a way as to create a systemic,
recursive processing loop. My hunch is that this would address
Aaronson's objections, as brains would score high on this measure
but the systems that Aaronson complains about, such as "systems
that do nothing but apply a low-density parity-check code, or other
simple transformations of their input data" would score low due to
lack of recursion.
Terren
On Tue, May 19, 2015 at 12:23 PM, meekerdb <[email protected]>
wrote:
On 5/19/2015 6:47 AM, Jason Resch wrote:
On Mon, May 18, 2015 at 11:54 PM, meekerdb <[email protected]>
wrote:
On 5/18/2015 9:45 PM, Jason Resch wrote:
Not necessarily, just as an actor may not be conscious in the
same way
as me. But I suspect the Blockhead would be conscious; the
intuition
that a lookup table can't be conscious is like the intuition that
an
electric circuit can't be conscious.
I don't see an equivalency between those intuitions. A lookup
table has a bounded and very low degree of computational
complexity: all answers to all queries are answered in constant
time.
While the table itself may have an arbitrarily high information
content, what in the software of the lookup table program is
there to appreciate/understand/know that information?
What is there is there in a neural network?
A computational state containing significant information content.
A lookup table has significant information content.
Integrated Information Theory makes some strides in explains this
I think:
http://en.wikipedia.org/wiki/Integrated_information_theory
http://www.scottaaronson.com/blog/?p=1799
Brent
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it,
send an email to [email protected].
To post to this group, send email to everything-
[email protected].
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it,
send an email to [email protected].
To post to this group, send email to everything-
[email protected].
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
--
You received this message because you are subscribed to the Google
Groups "Everything List" group.
To unsubscribe from this group and stop receiving emails from it,
send an email to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
http://iridia.ulb.ac.be/~marchal/
--
You received this message because you are subscribed to the Google Groups
"Everything List" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to [email protected].
To post to this group, send email to [email protected].
Visit this group at http://groups.google.com/group/everything-list.
For more options, visit https://groups.google.com/d/optout.
